climateprediction.net home page
Fedora 2 to Fedora 9 huge computation speed decrease

Fedora 2 to Fedora 9 huge computation speed decrease

Questions and Answers : Unix/Linux : Fedora 2 to Fedora 9 huge computation speed decrease
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile GrimmReaperSound
Avatar

Send message
Joined: 21 May 05
Posts: 13
Credit: 5,071,799
RAC: 98
Message 36235 - Posted: 27 Feb 2009, 1:33:09 UTC

I just did a major upgrade on the home server from Fecora Core 2 to Fedora Core 9. I\'m using the same motherboard and HD, but I kicked up the RAM to 2Gb.
I installed Boinc 5.2 and hooked up to the CPDN as before.
Now I\'m getting about 50% less computations with the same preferences as before.
Does anybody have an idea of what could be wrong? I\'m stumped.
Thanks
Michel
-----------------------------------------------
Always keep an open mind, even Einstein could not
explain how the universe worked,...
ID: 36235 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 36236 - Posted: 27 Feb 2009, 4:39:09 UTC

I see that you were using 4.19 before. This is why so many of your models crashed; they needed a 5.* version.
The latest version 5 is 5.10.45, but the project has moved on to version 6, which will be needed for graphics by most computers.

If by less computations you mean a lower benchmark, this may just be due to changes in the way BOINC calculates the benchmarks, which has changed several times since 4.19

50% less cpus running models would be what you\'d get if you went from a 5.2.* version of BOINC to a 6.* version, as the number of cpus is calculated differently. But you appear to be running version 5.2.13, and I don\'t think that there was a change in this from 4.19
N.B. There are now 2 options for setting the number of processors to use in a multi-core machine, depending on the BOINC version.

One thing to check, is that now it\'s possible to set preferences locally in the BOINC manager, and this is explicit for each computer.
If you were looking at the menu of the manager, and clicked OK to exit, (instead of Cancel) this would have created a preferences-over-ride file. If so, you can get rid of it by clicking the Cancel button in the manager menu preferences, and then clicking Cancel to get out of it.

At this point I\'m going to wait to hear from you again, as there are too many possibilities.


Backups: Here
ID: 36236 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2167
Credit: 64,403,322
RAC: 5,085
Message 36241 - Posted: 27 Feb 2009, 14:13:57 UTC

On October 28 you finished a hadcm3 at 4.47 s/TS.
On December 1, you finished a hadsm3 at 3.7 s/TS.
You currently have two models listed against that PC:
A hadcm3 running at 5.7 s/TS since it started, that hasn\'t returned a trickle since Feb 1st.
A hadsm3 started on Feb 9 running at 3.4 s/TS.

I\'m not sure what day you upgraded your computer, or how you are measuring the decrease in computations?
ID: 36241 · Report as offensive     Reply Quote
Profile GrimmReaperSound
Avatar

Send message
Joined: 21 May 05
Posts: 13
Credit: 5,071,799
RAC: 98
Message 36246 - Posted: 28 Feb 2009, 10:04:35 UTC

Clarifications:
I cleared the local pref as suggested by Les.
I updated my server on the weekend of Feb 7-8.
The speed decrease I was referring to is the \"Work Done\" displayed in the manager and used for all team/world rankings. Previously, the work done was always about the 185 mark. Now the Work done is sitting at the 38 mark.
I merged the 2 computer accounts 2 days ago since they were both for the same computer.

Les, is this the new benchmarking you where referring to? If it is, it\'s havoc on team/world rankings.

Geophi, Unfortunately I ran into some problems with my backup and was unable to recover the Boinc files from before the upgrade. So the previous models are toast. The s/TS has a lot of variations based on the model being worked on. This should not be a gauge of how much work is being done by a single computer since the model itself is the big variable.


Michel
-----------------------------------------------
Always keep an open mind, even Einstein could not
explain how the universe worked,...
ID: 36246 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 36247 - Posted: 28 Feb 2009, 14:08:38 UTC

\"Work done\" is what is called \"credit\" on the projects and stats sites.

On this project, it is dependent on trickles returned to the project.
Different types of models take different amounts of time to reach the point of returning a trickle, so the RAC, which is shown as \"Avg. work done\" in the manager, will vary.
Credit is different for each model type to reflect this difference.

\"Benchmarking\" is what is shown as such in the Messages tab, where it will appear about every 5 days. It\'s not used by this project to grant credit.
The results of this also appear on your server Account page for each computer, as
Measured floating point speed and Measured integer speed.


Backups: Here
ID: 36247 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 36250 - Posted: 28 Feb 2009, 14:51:07 UTC

The HADCMivolc Geo-engineering model still shows no trickles since 1 Feb. This is why your recent credit is so low; you\'ve only been getting credits from your other model. Is the HADCM advancing? What model year has it reached? Are its graphics normal?
Cpdn news
ID: 36250 · Report as offensive     Reply Quote
Profile GrimmReaperSound
Avatar

Send message
Joined: 21 May 05
Posts: 13
Credit: 5,071,799
RAC: 98
Message 36252 - Posted: 28 Feb 2009, 15:58:00 UTC

Les,
Thanks I didn\'t know that model variability was taken into account when assessing credits for the work.

mo.V,
Now that would explain why my average credit (or work done) is so low. Thanks.
Like I stated my backup failed, so all the Boinc related files where lost prior to Feb 7th. When I reconnected to CDPN, the HADCMivolc was never reassigned probably because my computer was now seen as a new computer. I have since merged the \"two\" computers back into a single.
How would I go about restarting the HADCMivolc model to allow it to work to completion without having any old files available?

On a side note: all the moderators here rock!
I\'m impressed with your speed of responses and your answers. I have friends who are running other Boinc projects and the support they get just makes me that much happier to be with CPDN.
Michel
-----------------------------------------------
Always keep an open mind, even Einstein could not
explain how the universe worked,...
ID: 36252 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 36254 - Posted: 28 Feb 2009, 18:00:52 UTC

Hi again Michel

Each type of model processes a different amount of data between trickles, so the credit per trickle needs to be different for each model type.

If the HADCM3ivolc model is there in your BOINC Manager Tasks tab you can continue to process it. If it\'s disappeared, you can\'t. The only way to get a lost or crashed task back and continue to process it is by restoring a backup made while the model was there and running. Anyway, if it\'s disappeared don\'t worry. I think your computer has one core and you have one task so that\'s enough until it completes.

If you want to make backups of the BOINC Data folder in future, you\'ll find a selection of methods in the CPDN READMEs. There\'s a link in my signature. I use Les\'s manual method which is quick and easy. But make sure that if you have BOINC 6 installed as a service you know how to completely exit from BOINC before backing up.
Cpdn news
ID: 36254 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2167
Credit: 64,403,322
RAC: 5,085
Message 36256 - Posted: 28 Feb 2009, 18:43:25 UTC

Since no trickles were posted from either model from 1 to 9 Feb, that will be factored into your RAC. The RAC is a weighted average over a 30 day period on CPDN, and the fact you had no trickles for at least 8 of the last 30 days plays into it. Your RAC should rise to about the same figures as before, as you continue to crunch over the next few weeks.

It\'s also possible that the merger of computers could have had an effect on your RAC. If so, that effect should also be gone in a few weeks.

On a side note, if you want to get the most work done per unit time on that computer, probably running hadsm3, hadsm3mh and hadam3 models would be best. Currently, the performance of hadcm3 in Linux is pretty dismal.
ID: 36256 · Report as offensive     Reply Quote

Questions and Answers : Unix/Linux : Fedora 2 to Fedora 9 huge computation speed decrease

©2024 climateprediction.net